Research interest
Explainable AI
Understanding what complex prediction models do, and how to explain their outputs in useful ways.
Explainable AI is a central theme in my research. I am interested in methods that describe what drives a prediction and in how to choose an explanation that answers the right question. My work includes Shapley values, counterfactual explanations, and tools for navigating different explanation methods.
The themes below connect publications, talks, software, and news across this site. This introduction is a starting point and can grow into a fuller account of the research.
Subthemes
Choose a topic to see the linked work below.
All workShapley valuesFeature contributions based on Shapley values.Conditional Shapley valuesShapley explanations that account for dependencies between features.CounterfactualsAlternative inputs that would change a model’s prediction.Data attributionHow training data contributes to model behaviour.Choosing XAI methodsMatching an explanation method to a particular question or need.
No linked work has this tag yet.